How to Use Chroma with Langchain: Simple Guide

To use Chroma with langchain, first install the chromadb package, then create a Chroma vector store instance with your documents and embeddings. Use this vector store in Langchain's retriever or chain to perform fast similarity search and retrieval.

Syntax

The basic syntax to use Chroma with Langchain involves creating a Chroma vector store by passing your documents and an embedding model. Then, you can create a retriever from this store to query similar documents.

• Chroma: The vector database instance.
• embedding_function: The model that converts text to vectors.
• documents: The list of text documents to index.
• retriever: Used to search the vector store.

python

from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings

# Initialize embeddings
embedding_function = OpenAIEmbeddings()

# Create Chroma vector store with documents
vectordb = Chroma.from_texts(texts=['doc1', 'doc2'], embedding=embedding_function)

# Create retriever
retriever = vectordb.as_retriever()

# Use retriever to get relevant docs
results = retriever.get_relevant_documents('query text')

Example

This example shows how to create a Chroma vector store with sample documents, then query it to find the most relevant document using Langchain's retriever.

python

from langchain.vectorstores import Chroma
from langchain.embeddings.openai import OpenAIEmbeddings

# Sample documents
texts = [
    "Langchain helps build LLM apps.",
    "Chroma is a vector database.",
    "Python is great for AI development."
]

# Initialize embedding model
embedding_function = OpenAIEmbeddings()

# Create Chroma vector store
vectordb = Chroma.from_texts(texts=texts, embedding=embedding_function)

# Create retriever
retriever = vectordb.as_retriever()

# Query the retriever
query = "What helps build applications with large language models?"
results = retriever.get_relevant_documents(query)

# Print the most relevant document
print(results[0].page_content)

Output

Langchain helps build LLM apps.

Common Pitfalls

Common mistakes when using Chroma with Langchain include:

• Not installing the chromadb package, causing import errors.
• Forgetting to provide an embedding function, which is required to convert text to vectors.
• Passing empty or invalid documents to Chroma.from_texts, resulting in no data indexed.
• Not calling as_retriever() before querying, which leads to attribute errors.

python

from langchain.vectorstores import Chroma

# Wrong: Missing embedding function
# vectordb = Chroma.from_texts(texts=['doc1', 'doc2'])  # This will fail

# Right: Provide embedding function
from langchain.embeddings.openai import OpenAIEmbeddings
embedding_function = OpenAIEmbeddings()
vectordb = Chroma.from_texts(texts=['doc1', 'doc2'], embedding=embedding_function)

Quick Reference

Step	Description
Install chromadb	pip install chromadb
Import Chroma and embeddings	from langchain.vectorstores import Chroma; from langchain.embeddings.openai import OpenAIEmbeddings
Create embedding model	embedding_function = OpenAIEmbeddings()
Create vector store	vectordb = Chroma.from_texts(texts, embedding=embedding_function)
Create retriever	retriever = vectordb.as_retriever()
Query retriever	results = retriever.get_relevant_documents(query)

Key Takeaways

Always provide an embedding function when creating a Chroma vector store in Langchain.

Use the retriever interface from Chroma to perform similarity searches on your documents.

Install the chromadb package before using Chroma to avoid import errors.

Pass meaningful text documents to index for effective retrieval results.

Call as_retriever() on the Chroma instance before querying for documents.